HardAI MLProductProduct case20 min to answer

How would you measure whether an LLM correctly executes a user’s requested actions?

Asked at

Google DeepMind

1Give yourself 20 minutes

2Answer out loud, not in your head

3Then compare with the answer below

Stuck? Show a way to structure it+

01Clarify the objective, user, constraints, and acceptable failure modes.
02Structure the response around task success, tool correctness, policy compliance, human evaluation.
03Compare credible alternatives and state the decision criteria.
04Finish with validation, monitoring, and what would change the decision.

Reference answer

strong answer

I would first define the outcome and acceptable error boundary. I would explicitly cover task success, tool correctness, policy compliance, human evaluation. I would select an approach only after comparing quality, latency, cost, safety, and operational complexity, then validate it with representative cases and production monitoring.

task success
tool correctness
policy compliance
human evaluation

Must be adapted to the interviewer’s scenario
Would benefit from concrete numbers and personal examples

alternative answer

A second valid approach is to begin with the highest-risk assumption and test it cheaply. I would explicitly cover task success, tool correctness, policy compliance, human evaluation. I would select an approach only after comparing quality, latency, cost, safety, and operational complexity, then validate it with representative cases and production monitoring.

task success
tool correctness
policy compliance
human evaluation

Must be adapted to the interviewer’s scenario
Would benefit from concrete numbers and personal examples

Grade yourself

task success

weight 30%

StrongExplains task success concretely and ties it to a decision or measurable outcome.

WeakMentions task success without implications, trade-offs, or evidence.

tool correctness

weight 25%

StrongExplains tool correctness concretely and ties it to a decision or measurable outcome.

WeakMentions tool correctness without implications, trade-offs, or evidence.

policy compliance

weight 25%

StrongExplains policy compliance concretely and ties it to a decision or measurable outcome.

WeakMentions policy compliance without implications, trade-offs, or evidence.

human evaluation

weight 20%

StrongExplains human evaluation concretely and ties it to a decision or measurable outcome.

WeakMentions human evaluation without implications, trade-offs, or evidence.

Then expect these follow-ups

What would your evaluation dataset contain?
Tests: depth
What if offline and online results disagree?
Tests: adaptability
How would you set the launch threshold?
Tests: adaptability

Hiring now

Product manager roles are open right now

landed applies to roles like these for you, on each company’s own site, with answers written for the role.

Get 5x interviews10 free auto-applies. No card.

Product Manager (UK Market)
autarc · Berlin
Senior / Principal Product Manager
Chrono24 · Remote
NBC News Product Manager
NBC News · New York, NEW YORK, United States

Or practice this question first

Next question